1 of 40

Machine Learning for Healthcare

Ali Akbar Septiandri

@aliakbars

2 of 40

Outline

  • What is machine learning (ML)?
  • Opportunities in machine learning for healthcare (ML4HC)
  • Challenges in ML4HC
  • Future of ML4HC

3 of 40

What is machine learning?

4 of 40

How to identify cats or dogs in an image?

5 of 40

Identifying Cats or Dogs

Model

  • Logit model
  • SVM

Image Processing

  • Edge detection
  • Texture analyser
  • Color histogram

Feature Extraction

  • Eye position
  • Eye colour
  • Nose colour
  • Fur type
  • Leg counts

6 of 40

y = σ(β0 + β1x1 + β2x2 + β3x3)

Logit model from defined features

7 of 40

“Fundamentally, machine learning involves building mathematical models to help understand data.”

  • Jake VanderPlas

8 of 40

…but it might not even work!

9 of 40

Enter: Deep Learning

10 of 40

(Artificial) Neural Networks ~ Deep Learning

11 of 40

It’s that simple*!

Output

Feature extraction

+ model training

Input

12 of 40

“No free lunch”

13 of 40

Opportunities

14 of 40

Verbal screening using mobile app

15 of 40

Stanford ML Group

16 of 40

17 of 40

“...it could diagnose 14 diseases with 80% accuracy–in other words, about as well as a real radiologist.”

18 of 40

Deep Neural Networks Improve Radiologists’ Performance in Breast Cancer Screening (Wu et al., 2019)

19 of 40

20 of 40

21 of 40

22 of 40

Making out of focus microscopy images in focus again

23 of 40

Early Warning System

System for Electronic Notification and Documentation (SEND) by Sensyne Health

24 of 40

Natural Language Processing (NLP) for clinical notes

25 of 40

DeepMind’s AlphaFold

26 of 40

“...diagnosing and treating diseases believed to be caused by misfolded proteins, such as Alzheimer’s, Parkinson’s, Huntington’s and cystic fibrosis.”

(Evans et al., 2018)

27 of 40

Challenges

28 of 40

Accuracy

29 of 40

30 of 40

“Correlation does not imply causation” | Image from xkcd

31 of 40

Understanding Causality

Graphical presentation of confounding in directed acyclic graphs (Suttorp et al., 2014)

32 of 40

Missing Data

33 of 40

Adversarial Attacks

34 of 40

Bias in ML

“...a project to look for skin cancer in photographs. It turns out that dermatologists often put rulers in photos of skin cancer, for scale, but that the example photos of healthy skin do not contain rulers. To the system, the rulers (or rather, the pixels that we see as a ruler) were just differences between the example sets, and sometimes more prominent than the small blotches on the skin. So, the system that was built to detect skin cancer was, sometimes, detecting rulers instead.” (Evans, 2019)

35 of 40

Future of ML4HC

36 of 40

Optimal treatment strategies, e.g. for sepsis (Komorowski et al., 2018)

37 of 40

38 of 40

39 of 40

Where are we now?

40 of 40

Thank you.

aliakbars@live.com

@__aliakbars__

http://uai.aliakbars.com